Short-Term Traffic Flow Forecasting By Fusing Spatial- Temporal Traffic Information

As a sophisticated and intelligent traffic management mode, the realization foundation of traffic flow guidance and control system is real-time, dynamic and precise short-term traffic prediction. City road network traffic flow has a tight space-time logic relation, in theory, only fully integrating the space-time movement information of the prediction point and its upstream and downstream related point can we accurately simulate the true motion characteristics of traffic flow. At the same time, considering the factors of history or reality, many cities cannot provide long-term traffic flow history database, So the short-time traffic flow forecasting model is need to be established under the condition of a small sample data. In this paper, the research work is based on above two points.Firstly, this thesis summarizes the current mainstream traffic flow forecasting model, and points out that the hybrid model based on fusion theory and the spatiotemporal composite prediction model is the focus of the current research of traffic flow forecasting. Secondly, this article elaborates the spatiotemporal relationship between network traffic flow, gives the prediction principle by fusing spatio-temporal information, clearly makes the prediction point and its direct connection point as the system analysis object. Next, in order to reduce the input dimension, filtering invalid information, the grey comprehensive correlation degree method is used to screen out the strong association points from direct association points. The improved grey Elman neural network prediction algorithm is constructed, combining the grey theory with Elman neural network, aiming at the problem of Elman network whose identification ability for high order system is insufficient, the network structure is improved, strengthen to reflect the dynamic nature of traffic flow. The improved Niche PSO-GRNN prediction algorithm is constructed, introducing the niching technique and ameliorating the inertia weight and learning factors, get rid of the drawbacks of the traditional PSO algorithm "local optimal solution" and "premature convergence", the improved NPSO algorithm is used to establish the optimization model for the smooth factor of GRNN model.Through simulation experiments, it can be found that under the condition of small sample data, two kinds of improved models proposed in this paper both show good prediction performance, in the forecasting index, convergence speed and generalization ability are better than that of their original model. Finally, prediction accuracy and prediction stability of the model are comprehensively considered, it is therefore concluded that the improved grey Elman neural network model has higher engineering practical application value than the improved NPSO-GRNN model.